US Appeals Court Rejects Fair Use Defense in Landmark AI Training Case
In the first US appeals court decision on AI training, the Third Circuit ruled that copying proprietary headnotes to build a competing search tool infringes copyright.
Key takeaways
- The 3rd US Circuit Court of Appeals upheld a summary judgment ruling for Thomson Reuters against defunct startup ROSS Intelligence.
- The court ruled that training an AI legal search engine on copyrighted Westlaw headnotes was minimally transformative and did not constitute fair use.
- This decision marks the first time a US federal appeals court has ruled on fair use defenses concerning AI training datasets.
- The panel found Westlaw's headnotes met the legal threshold for originality and that ROSS's platform acted as a direct commercial market competitor.
In a milestone decision for artificial intelligence and copyright law, a federal appeals court has rejected an AI developer's fair use defense for training models on copyrighted content. The Philadelphia-based 3rd US Circuit Court of Appeals affirmed a lower court ruling in favor of information services company Thomson Reuters against former legal-research rival ROSS Intelligence, as reported by CNA.
The ruling marks the first time a US circuit court has directly addressed whether training artificial intelligence systems on copyrighted material qualifies as fair use. The panel concluded that extracting proprietary legal summaries to train a commercial AI-powered search tool was not transformative and infringed on Thomson Reuters' intellectual property.
The Roots of the Legal Dispute
Thomson Reuters first filed its lawsuit against ROSS Intelligence in 2020. The complaint accused ROSS of copying more than 2,000 copyrighted "headnotes"—concise editorial summaries of legal points appearing before judicial opinions—from Thomson Reuters' Westlaw platform, according to IPWatchdog.
After Westlaw refused ROSS direct access, the startup obtained the data through a third party, LegalEase Solutions. ROSS used the headnotes to train an AI-based search engine designed to return relevant judicial opinions to user queries. Facing mounting legal expenses from the litigation, ROSS shut down its platform in 2021.
In Delaware federal court, US Circuit Judge Stephanos Bibas, sitting by designation, issued a revised summary judgment ruling in 2025 finding that Westlaw's headnotes and Key Number System taxonomy possessed the necessary originality for copyright protection. Judge Bibas rejected ROSS's fair use defense, prompting ROSS to appeal to the Third Circuit in June 2026.
Why the Third Circuit Rejected Fair Use
Writing for the appellate panel, US Circuit Judge Tamika Montgomery-Reeves—joined by Judges L. Felipe Restrepo and Emil J. Bove III—clarified that despite the technological framing, the dispute rested on core copyright principles, as detailed by Courthouse News Service.
"Under ROSS' framing, this case appears to concern the future of AI legal technology," Judge Montgomery-Reeves wrote. "But appearances can be deceiving. In truth, this is no more than an ordinary copyright case."
The panel evaluated the originality of Westlaw's headnotes, determining that editors exercised a creative spark in drafting summaries that stood independently while accurately reflecting judicial rulings. Addressing the fair use defense, the court described ROSS's use as "minimally transformative at best."
"ROSS took an intermediate step of using the headnotes to train an AI program, which arguably presents a slight degree of difference in use," Montgomery-Reeves stated in the opinion. "But the undisputed evidence demonstrates that ROSS used the headnotes to train an AI program for the benefit of its legal-research platform."
The judges also determined that ROSS's system would harm Westlaw's commercial value in the legal research market as well as its potential licensing market for AI training data.
Broader Implications for AI Training Lawsuits
A spokesperson for Thomson Reuters stated the company was pleased with the decision, emphasizing that "respecting copyright is essential for fostering innovation while protecting the intellectual property that fuels fiduciary-grade AI solutions," according to CNA.
While the Thomson Reuters case focused on a non-generative search engine rather than generative models that synthesize new text or imagery, the appellate precedent arrives amid dozens of active lawsuits brought by authors, news organizations, and music publishers against major AI developers.
The Third Circuit issued its initial order under seal and instructed the parties to submit proposed redactions within 10 days before the public release of the complete opinion, according to IPWatchdog.
Frequently asked questions
What did the Third Circuit rule regarding AI training and fair use?
The Third Circuit held that ROSS Intelligence's unauthorized use of Thomson Reuters' Westlaw headnotes to train an AI legal search platform was minimally transformative, competed directly with Westlaw, and did not qualify for fair use protection.
Did this lawsuit involve generative AI?
No. Unlike lawsuits involving large language models that generate text or art, this case involved a non-generative legal search engine that used headnotes to identify relevant judicial opinions.
Why were Westlaw's legal headnotes deemed copyrightable?
The court found that Thomson Reuters' editors exercised a sufficient 'creative spark' and original judgment in summarizing complex judicial opinions into self-contained legal headnotes.
Sources
- US appeals court upholds Thomson Reuters' landmark win in AI training lawsuitCNA · Sep 29, 2026
- AI training of copyrighted material not fair use: Third CircuitCourthouse News Service · Sep 30, 2026
- Third Circuit Affirms Revised Fair Use Ruling Against ROSS’ AI Legal Research Platform in Sealed OpinionIPWatchdog.com | Patents & Intellectual Property Law · Sep 30, 2026
How this story was made: the newsroom picked it up from Techmeme, Reddit and Google News, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (29 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.
Published October 1, 2026 at 08:05 UTC
